From Claude Code in your terminal to a live site
Yes, Claude Code can deploy a website: one claude mcp add command, a sign-in with /
Short answer
Yes. Claude Code is an MCP client, so deploying is one more tool it can call. Run claude mcp add --transport http --scope user vibedeploy https:// once, start a session, run /mcp and sign in to VibeDeploy in the browser. Then say "deploy this": Claude Code uploads the project and prints the live URL. An app that needs a server or a database runs in containers next to the site.
Ship it from the terminal
One command connects Claude Code to VibeDeploy. After that, "deploy this" puts the project on a live URL and your own domain. 14 days free, no card.
claude mcp add --transport http --scope user vibedeploy https://mcp.vibedeploy.be/mcpCan Claude Code deploy a website?
Yes. Claude Code reads your codebase, edits files and runs commands, and it is an MCP client, so a deploy is just another tool it can call once a host is connected. Connect VibeDeploy with one command:
claude mcp add --transport http --scope user vibedeploy https://mcp.vibedeploy.be/mcpThen start a session, run /, pick vibedeploy and sign in in the browser window that opens. From then on, "deploy this" is the deploy: Claude Code uploads the project and prints a live URL of the form name.subsite.site. If the project needs its own server or a database, the same conversation can start those as containers next to the site.
The rest of this post goes deeper: why Claude Code changes the deploy story, the two ways to make it ship its own work, what happens to projects with a backend, and the loop that turns your terminal into the only tool you touch.
Where a Claude Code session actually leaves you
Claude Code is different from the chat tools in one way that matters here: it writes files directly into your project folder. There is nothing to copy out of a chat, no preview to escape, no download step. When a session ends, the site is already on your disk, complete.
You can run a local preview and it looks finished. It is finished, as code. But a local preview is visible to exactly one computer on earth, and a folder is a website the way a printed manuscript is a published book. The gap is hosting, and the usual way to cross it, push to GitHub, sign up for a separate deploy platform, wire up a pipeline, is a second job the agent just spent an hour helping you avoid.
The better move is to make the agent close its own loop. Claude Code is both a coding agent and a shell, so it is capable of doing the deploy itself. There are two routes, and both end in a live URL printed in the same terminal window where the code was written.
Route one: MCP, the smooth road
Claude Code is a native MCP client, which means it can connect to a deploy server and gain deploy actions as first-class capabilities. The command above is all it takes, run once. --scope user makes the server available in every project on your machine; without it, the server registers for the current project only. Stay on the http transport, since the older SSE transport is deprecated.
Sign in with your VibeDeploy account when / opens the browser window. No account yet? The sign-in window can create one on the spot (a 14-day trial, no card), or you can start the trial first. That sign-in is the last time you leave the terminal. From then on the deploy is a sentence: "deploy this site to VibeDeploy." The agent calls the deploy tool, uploads the files, and prints the live URL. The Claude Code guide and the MCP setup page have the same steps with copy buttons.
If your project is a shared repository, you can commit the connection so it travels with the code rather than living on one machine. Drop a .mcp.json at the repo root:
{
"mcpServers": {
"vibedeploy": {
"type": "http",
"url": "https://mcp.vibedeploy.be/mcp"
}
}
}Keep the "type": "http" line. Claude Code reads an entry without a type as a local stdio server and skips it. With the line in place, everyone who opens the project in Claude Code gets the same deploy target and signs in with their own account. The deploy capability becomes part of the repo, not a per-developer chore.
Route two: the deploy guide, no MCP at all
MCP is the comfortable path, but it is genuinely optional, because Claude Code can run shell commands. That means it can talk to VibeDeploy's plain HTTP deploy API directly. Copy the deploy guide from your dashboard (it is labelled Magic File there), put it in the project folder, and tell the agent to deploy. It reads the instructions, packages the files, and POSTs them to the API itself. No MCP setup, no extra command, just a file the agent knows how to follow.
The guide contains a private deploy key, so keep it out of Git: add it to .gitignore, or store it outside the repository and point the agent at it. This route matters in two situations: a locked-down environment where you would rather not register an MCP server, and a first deploy where you want the shortest possible path before committing to a setup.
Ship it from the terminal
One command connects Claude Code to VibeDeploy. After that, "deploy this" puts the project on a live URL and your own domain. 14 days free, no card.
claude mcp add --transport http --scope user vibedeploy https://mcp.vibedeploy.be/mcpWhich projects deploy, and what happens to a backend
Static output is served directly, so the first question is what your Claude Code project compiles down to:
- Plain HTML, CSS and JavaScript. Deploys as it is. Nothing to build.
- Vite, Astro, or a static Next.js export. Ask Claude Code to run the build first; it knows the command. The
dist(orout) folder is what deploys, and clean URLs work, so/is answered byabout about.html. - A single-page React or Vue app. Unknown paths fall back to
index.html, which is what client-side routing needs. A site with real pages should ship a top-level404.htmlso wrong URLs return a real 404.
A project that needs a running server, such as server-side rendering, API routes or a live database connection, does not have to move to another platform. Static sites and single-page apps are served directly. A site can also run containers next to it (your own server, Postgres, MySQL, Redis or workers; up to 6 services with 2 CPU and 2 GiB in total), deployed by your AI through the MCP tool deploy_app and isolated with gVisor. Container data is not yet included in snapshots.
In practice you ask Claude Code something like "run this app's API and a Postgres database next to the site". It writes the service definition, starts the containers, and reads their logs with get_app_status when something fails. Two rules come with it. API keys and OAuth client secrets go into the dashboard, in the site's Settings tab under Secrets, never into the terminal session. And there is no dashboard view of container logs yet, so the agent is how you see them.
Anthropic draws the same line for its own feature. Claude Code can publish a session's output as an artifact on claude.ai, but its documentation describes an artifact as one self-contained page with no backend, and says a hosted internal tool with a backend belongs on your own infrastructure.
The redeploy loop is the real win
The first deploy is nice. The second one is the point. Once the connection exists, your workflow collapses to a single window: describe a change, Claude Code edits the files, Claude Code redeploys, the URL serves the new version. No context switch to a browser, no commit-and-wait, no separate deploy dashboard.
Redeploying costs nothing extra. The price is flat per plan, with no per-deploy fee and no bandwidth overage billing; each plan does cap the number of deploys per hour. Every deploy is also kept as a version with a restore point taken before it, so if a revision breaks something, the previous version comes back with one click in the site's History tab in the dashboard, within your plan's restore window. That restore is a dashboard action: the MCP server has no restore tool, so a bad deploy is undone by you, not by the agent.
This is the same iteration loop that makes the chat tools productive, but tighter, because the terminal never has to hand control to anything else. The agent that wrote the line also shipped it.
Putting it on your own domain
A name.subsite.site address is right for testing. For anything public, attach your domain, or let Claude Code do it: ask it to add your domain to the site and it calls add_custom_domain, which returns the DNS records to publish. That is a TXT record that proves the domain is yours, and a CNAME to proxy.vibedeploy.be, or an A record for an apex domain. Once you have published them, verify_custom_domain checks the records, and the certificate is issued automatically as soon as DNS points at VibeDeploy. Until then the site stays reachable on its subdomain. From then on the custom domain serves every redeploy the agent makes, with no further DNS work.
What it costs, and where to go next
The 14-day trial covers one site with a custom domain you already own and a container next to it, so the whole flow above can be tried end to end without a card. After that, Solo is EUR 4 a month, Maker EUR 15 and Studio EUR 39, and prices include VAT where it applies. The tiers are on the pricing page.
Coming from Bolt.new? Move a Bolt.new project to Claude Code brings the code over and keeps the site live along the way. If your Claude work happens in the chat rather than the terminal, the artifact route is its own guide, Claude artifacts to your own domain, and the map of every AI tool and deploy route is the complete guide to deploying AI-built websites. But if you live in Claude Code, the shape of the answer is simple: the terminal that built the site can be the terminal that ships it, and the round trip from idea to live URL never leaves the window.
Sources
- Claude Code docs, Overview: what Claude Code does (reads your codebase, edits files, runs commands) and where it runs. Checked 13 October 2026.
- Claude Code docs, Connect Claude Code to tools via MCP:
claude mcp add --transport http, the local, project and user scopes, signing in with/, themcp .mcp.jsonformat and the warning that an entry without a type is read as a stdio server, and the deprecated SSE transport. Checked 13 October 2026. - Claude Code docs, Share session output as artifacts: artifacts as single pages on claude.ai with no backend, and the advice to deploy a hosted tool with a backend on your own infrastructure. Checked 13 October 2026.
Ship your AI-built site in minutes
VibeDeploy hosts your AI-built websites in the EU with custom domains, automatic SSL, and a 14-day free trial that gets you online today.
Deploy guides for this topic
Related reading
"Uploads the file below from this task to claude.ai": Claude's Publish Artifact Prompt Explained
What Claude's Publish artifact prompt means: "Uploads the file below from this task to claude.ai", who can see the page, the three variants, and how to undo it.
Claude Artifact Custom Domain: Download, Host, and Launch
Download or copy your Claude artifact code, host it outside Claude, and connect a custom domain with EU hosting and automatic Let's Encrypt SSL included.
How to Deploy AI-Built Websites in 2026: The Complete Guide
Every way to get an AI-built site live in 2026: MCP deploys, a deploy guide for any chat AI, upload or GitHub, containers, and a 25-tool export matrix.
Move a Bolt.new Project to Claude Code (and Keep It Live)
Move a Bolt.new project to Claude Code: sync it to GitHub or download the ZIP, add a CLAUDE.md, connect VibeDeploy and deploy to your own domain.